Face Forgery Detection with Long-Range Noise Features and Multilevel Frequency-Aware Clues

Figure 2

Visualization examples of frequency decomposed components for the real and four manipulation scenes in the FF++ [20] high-compression subdatasets. Capturing the differences between the real and forged images is challenging within the low-frequency component. Two face-swapping methods expose subtle differences in facial textures within the middle-frequency component. Meanwhile, four tampering methods expose subtle manipulation traces in different facial regions within the high-frequency component.